It is now one function. In later commits we will extend this by checking helper
segments separately, which should also be more efficient.
It changes one test-case, probably because ofn ear_offsetted. The change is not visible
and the result is probably more correct. SQL Server reports this polygon as invalid.
In the past there was one with a straight line going from one piece to the next.
Now there are two, going from one piece to the point on the geometry, and the
second going from the point on the geomety to the next piece.
They are now zero-volume.
They fix:
1) crossings of sides with flat-ends (unit-test cases curve, field_sprayer, bend_near_start)
2) joins with limited number of points (unit-test cases concave_b_rough, italy_part, nl_part)
3) joins with large distances
4) many self-intersections on remaining aimes-cases
features in the original (convex/concave/convex) being enlarged
in a very large scale and causing issues (IP's within pieces).
This might be reconsidered later. Simplifying with a very small
distance will never be visible in the result,
if it is using round joins. For miter joins they are even more
sensitive to small scale input features, however the result will
look better.
A distance strategy can change behaviour (make larger - set simplify
distance to 0.0 to avoid simplification)
This fixes the parcel issue(s)
This moves responsibility of generation to the calling code, removes it
from the join strategies. It also enables using the end-strategy for the
spikes
there was no segment in between generated, therefore if the side-segments
did not intersect (e.g. donut_diamond in polygon_buffer.cpp unit test),
there was no intersection at all. Now a helper piece is generated and it
will intersect.
